Educational applications based on diagnostic learning analytics model in the context of big data analytics

Under the context of big data analytics, Diagnostic Learning Analytics (DLA) offers a novel approach to understanding and improving student learning behaviors. This study proposes a DLA model that integrates the M-Canopy-K-means clustering algorithm with the PSSHRNet-based pose estimation network to perform deep mining and analysis of classroom behavior data. The experimental results show that the proposed M-Canopy-K-means algorithm improves the clustering accuracy by 10 % and 7 % respectively compared with the K-means and Mini-batch K-means algorithms, and achieves an speedup ratio of more than 2 times on small-scale datasets. In behavior recognition, the PSSHRNet-based network achieved lower loss values and faster convergence on MSCOCO and CrowdPose datasets. In practical applications, the model detected that a student with high engagement had an actual participation score of only 2.81, with mobile phone use accounting for 33.29 % of classroom behavior. The study demonstrates the effectiveness of big data-driven DLA models for diagnosing educational behaviors and provides a solid foundation for personalized teaching interventions.